dc.description.abstract | Finite element method (FEM) simulations are widely used to perform structural analysis. To ensure that the results obtained from the simulations are realistic it is crucial to have a material model that is as close to reality as possible.
This thesis presents a new method for identifying material parameters based on artificial neural network (ANN), Genetic Algorithm (GA), numerical simulations and experimental tests. The method is based on a classical inverse identification procedure aiming to find the material parameters, minimising the difference between the simulated response by FEM analysis and the measured response from an experiment. However, to save computational time the FEM simulations are substituted by an ANN in the optimisation loop.
In order to test the method, two case studies were carried out with the purpose to identify the parameters in the Chaboche hardening model and in the Gurson Tvergaard Needleman (GTN) model. The results from the case studies showed that the method was able to provide relatively accurate results using a short amount of computational time. Furthermore, in both cases it was observed that different set of material parameters could give a nearly identical load-displacement curve. It is therefore recommended to extend the method so that it uses more quantities in the comparison between the simulated and experimental result. However, by making this adjustment it is expected that the method can outperform conventional methods when computational time and accurate results are both emphasised. Therefore this strategy should be considered in the future when the optimal material model is to be determined. | |